ADS127L01: 24bit resolution?

Hello Raghu,

ADC manufacturer's classify an ADC resolution by the number of bits that you can read from the device.  ENOB is based on the internal noise of the ADC and the resolution.  Different ADCs have different internal noise levels, which is why the ENOB for a given data rate is different for various ADCs.

Even though the ENOB is less than 24b, you can always use additional digital filtering on the collected data to increase the ENOB of the measurement.  As an example, the ADS127L01 is specified with ENOB(or effective resolution) of 21.83b (1.34uVrms noise) in VLP mode and OSR=2048.  If you further filter this data in the host processer, you can increase the ENOB.  Using a simple average filter and averaging 16 readings, the noise will reduce by a factor of 4, Vnoise=1.34uVrms/sqrt(16)=0.335uVrms.  This will result in an ENOB(effective resolution) of 23.83b.

You may want to take a look at the TI Precision Labs -ADCs.  There is a lot of good information in these presentations that discusses noise in both SAR and Delta-Sigma ADC topologies.  Specifically, section 4, Analog-to-digital convert (ADC) drive topologies has some good information about this topic.

Regards,
Keith Nicholas
Precision ADC Applications

Hi Raghu,

With delta-sigma ADC's, there are usually multiple filter options, so depending on the specific filter used, the effective resolution will change.

You are correct, the amount of noise reduced on the digitally filtered data will depend on the filter response.  However, 1/sqrt(OSR) is a good approximation.  The only way I know to get an exact number is to mathematically combine the filter response of both filters (the internal digital filter and the external filter response resulting from averaging in this example), and then integrate the overall response. 

Here is an example from the new ADS127L11 delta-sigma that has an option for a SINC4+SINC1 filter.  For the lower SINC1 OSR (number of averaged results from the SINC4), the reduction in noise is predicted by 1/sqrt(OSR), but for higher levels, the reduction is not as great, but the simple equation predicts the total noise for OSR up to 100 to within 24%.

SINC4 filter, OSR32, Noise=7.96uVrms

SINC4, OSR32 plus SINC1, OSR2, Noise=7.96uVrms/sqrt(2)=5.63uVrms, actual noise 5.63uVrms, 0% error

SINC4, OSR32 plus SINC1, OSR100, Noise=7.96uVrms/sqrt(100)=0.796uVrms, actual noise 0.99uVrms, 24% error

Regarding the use of the term ENOB in TI's older delta-sigma datasheets, the resolution term that we are specifying is effective resolution, not ENOB.  All of the above examples are effective resolution, which is a DC specification based on the internal noise of the ADC when the inputs are shorted.  This can be calculated using the following formula:

Effective Resolution (bits) = 3.32*log₁₀(FSR/noise)

FSR=full scale input range, for the ADS127L11 in 1x range, this is 2*Vref.

ENOB is based on AC performance, where a low noise and low distortion sinewave (1kHz is common) is sampled, run through an FFT algorithm, and the SINAD (ratio of full-scale signal-to-noise and distortion) are calculated and used in the following formula:

ENOB=(SINAD(dB)-1.76dB)/6.02

ENOB will always be less than effective resolution, since the FSR is based on the maximum RMS level of a sinewave, or FSR/[2*sqrt(2)]

If the ADC distortion is very low and the noise is dominated by thermal noise, then you can estimate ENOB=Effective Resolution - 1.8b.

On new device datasheets, we are using the term effective resolution if it is based on input shorted noise.  If ENOB is used, then it is based on the above equation using an input sinewave.

I hope this helps!

Regards,
Keith

Hi Raghu,

Just to be clear, the SINC1 filter is equivalent to an averaging filter.  So in the above example, the SINC4 output is averaged by the SINC1, where OSR is the number of samples from the SINC4 output that are further averaged together.

In the above table, the SINC4 filter is fixed at OSR32; it uses 32 modulator input samples to generate an output that has a data rate (internal to the ADC) of 400ksps.  The SINC1 takes the SINC4 output and further averages these values where OSR is equal to the number of output samples from the SINC4 that are averaged together.  If you take the output of the ADS127L01 low-latency filter and further average it in your host processor, you can expect to get further reduction in noise, which will result in higher effective resolution (and ENOB).

The advantage of breaking the filters up is a tradeoff between noise and latency.  Using the ADS127L11 again as an example, setting the SINC4 to an OSR of 128 results in an output data rate of 100ksps (high-speed mode, Fclk=25.6MHz), a noise of 3.90uVrms and a latency of 40.63usec (total time for the digital filter to settle).  Using the SINC4+SINC1 option, where the SINC4 OSR is fixed at 32 and the SINC1 is set to an OSR of 4, you will get the same 100ksps data rate, but the noise has increased to 3.98uVrms and the latency is now 18.13usec.

Regards,
Keith