This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

ADS1294R: LF-Noise of the integrated reference.

Part Number: ADS1294R
Other Parts Discussed in Thread: REF5025, REF2125, ADS1294


First the fundamental problem: I run a battery-driven system (very compact) with the ADS1294R for respiration detection at effectively 10Hz. In order to reduce the power consumption I turn the ADS on when a measurement has to be done and then off again. This only works with an external reference because the internal one requires at least 150ms startup time.

The external reference shall also not draw too much power but shall provide a noise level low enough for a sufficient measurement - the critical part is the 1/f-noise at freq. <1Hz. ADS129xR datasheet tables 1 and 2 provide noise information for at least 1k samples - but for 32ksps this means only 31ms and for 500Hz a 2sec measurement period. Hence, this information is not suitable if I want to go into respiration detection.

Of course, before carrying out first tests with an external reference I also used the internal one which - surprise, surprise - did lead to good results.

What I'd like to have is a simple information regarding the noise level for 0.1-10Hz and e.g. 10Hz-10kHz for the internal reference to have a data basis for a proper external reference circuit design.

Thank's in advance.

  • Hello Jurgen,

    Thank you for your interest in our ADS1294R!

    I can ask our designers if we have more specific information on the noise of the internal reference. Have you also looked at Figure 3 in the Typical Characteristics table? There we show the peak-to-peak noise of a single channel captured over a 10-second period. The data rate and gain settings are listed in the heading above. As you can see, the measured noise in this configuration is within 5uVpp with the ADC inputs shorted.

    Best Regards,
  • Good Morning!

    Figure 3 with 500 sps means that it may have noise with at least 250Hz - if I had the raw data for this figure I could extract the 0.1-10Hz noise but the figure itself does unfortunately not provide that information. However, looking to it seems to show a low-frequency-noise with 1.5 to 2uVpp - what is the fraction of the input amplifier and what is given by the reference? Pls. note that the figure caption provides no information regarding the utilized capacitors at VCAP1 (22uF?) and VREFP-VREFN (10uF?).

    Something else regarding the recommended ext. reference at Fig 32: REF5025 has a typ. noise level of 3uVpp/V, i.e. 7.5uVpp for 2.5V. Wouldn't it be more suitable to claim REF2125 and make the LPF following the ref-output at least with 1k and 100uF? Or is that filter only supposed to suppress HF- and residual supply noise?

  • Hi Jurgen,

    The noise you see in the plot should only come from the thermal noise in the PGA and delta-sigma modulator as well as the ADC's quantization noise. The reference noise will not contribute to this plot as the measurement is done with the inputs of the PGA shorted together and biased to mid-supply. Reference noise only plays a role with non-zero inputs and increases for larger input voltages.

    After discussing this with the designer, I wanted to share additional details on the internal reference start-up time. The 150ms is predominately limited by the RC time constant between R1 and the capacitor to ground on the VCAP1 pin. For R1 = 12.5k and C = 22uF, one tau is equal to 275ms; however, we spec a typical 150ms since this is typically the time required by the internal bandgap voltage in order to read correctly from the device programmed registers. You can actually reduce this time by scaling the value of the external capacitor. While this cap does provide additional filtering, the loss in noise performance will not be detrimental if you were to scale it down by 10x or even 100x (especially true since reference noise only dominates at large differential input voltages to the ADC). The external cap between VREFP and VREFN would stay as-is since it is needed to provide charge to the internal reference sampling circuit.

    The lower limit to which I would scale this capacitor is set by the device's internal tPOR (power-on reset) timing, equal to 2^18 tCLK. For CLK = 2.048 MHz, tPOR is at least 128ms (or longer if VCAP1 is still < 1.1 V). Therefore, it only makes sense to reduce VCAP1 to 10uF. This would allow you the most filtering of the internal bandgap voltage without extending the required start-up time any further than required by tPOR.

    Personally, I would suggest to go with this approach rather than design your own external reference.

  • Thank's for your detailled answers. After playing around with a few equations I think it is possible to state "the ratio of the reference-noise to the nominal reference voltage has to be smaller than the ratio of the peak-peak-signal to the signal's base line", i.e. for bioimpedance with delta_R=0.1Ohm and baseline R0=1kOhm the ratio is 100ppm which means the peak-peak-noise of the reference has to be smaller than 100ppm*2.5V=250uV - for 20dB SNR I would then need a reference with no more than 25µVpp noise (0.1-10Hz) - am I right?

    BTW: The 2^18 clock cycles only apply for a full reset, but I power the ADS down with the respective input pin and turn it on again (followed by re-writing the register map) - that only takes a few tens of Microseconds (see my other post) - so I still need an external fast reference ;(
  • Hello Jurgen,

    Understood. I wanted to confirm with you that, during /PWDN = 0, are both analog and digital supplies still present? Will your external reference voltage remain present while the ADS1294 is in /PWDN?

    If the supplies remain present, then following /PWDN = 1, you are correct that you do not need to follow the tPOR procedure. In addition, the register map will retain the previous settings, so you do not have to re-write them.

    I'm not entirely sure on the rule of thumb you presented with regards to external reference noise. In theory, it does make some sense, but remember a couple things:

    1. The digital filter will further limit your effective noise bandwidth of the signal chain. You mentioned earlier that at 500SPS, you could effectively see noise up to 250Hz. This is true, but the SINC3 digital filter will already attenuate that noise by 3dB at 0.262*fDR, so the contribution of this noise becomes even less.
    2. For small input signals (i.e. VIN << VREF), the noise contribution from the voltage reference is much more negligible.

    It looks like either the REF5025 or the REF2125 would work for your application, but note that the REF2125 has 2x higher drift (6ppm compared to 3ppm) and 67% higher noise (5uVpp/V vs.  3uVpp/V). I still would not recommend using a large resistor in series with the reference output and VREFP without following it with a buffer.

    Best Regards,

  • Hi!

    Ad 1: You will agree that even a decimation filter is not able to filter noise that arises from the reference - that's why we call it 'reference', don't we?

    Ad 2: This is what my simple description means. My concerns regarding noise is mainly driven by the 1/f-noise: For breathing detection we talk about at periodicity of e.g. 3 to 10 seconds (depending on activity). For very low breathing rates (i.e. during resting) the additional challenge is that breathing mainly occurs via the diaphragm but the chest/ribcage does not contribute that much and the signal change is very small. Temperature will also change when applying a device but this change is much slower than a few breathing cycles and the absolute values >1minute ago are not of interest.

    To clarify my statement regarding noise: Imagine a signal with a certain baseline Vcm with an alternating (peak-peak) signal Vs,pp on top and a reference Vref with a noise (peak-peak) Vn,pp. As the worst-case scenario I want the binary converted signal still shows that (Vcm+Vs,pp/2) is larger than (Vcm-Vs,pp/2) even for the worst-case noise, i.e.

    (Vcm+Vs,pp/2)/(Vref+Vn,pp/2) > (Vcm-Vs,pp/2)/(Vref-Vn,pp/2)

    To simplify things, one can define two variables alpha=Vs,pp/2/Vcm and beta=Vn,pp/2/Vref. Solving that unequation then leads to alpha>beta which is represents what you considered: For small peak-peak-values on top of a small baseline alpha is quite large, so beta (i.e. the reference noise level) may also be 'larger'.