ADS1115/Getting a good number out of it/07. Data rate, and the bits that are noise
Getting a good number out of it · 07 of 11

Data rate, and the bits that are noise

Eight conversions a second or eight hundred and sixty, chosen with one register. The fast settings do not give you sixteen bits sooner — they give you fewer bits, because the converter has less time to average, and TI prints exactly how many you lose.

Where the quiet comes from

This is a delta-sigma converter, which means it samples its input far faster than it reports — 250 000 times a second internally — and then averages those samples down to whatever output rate you asked for.

So the rate setting is really an averaging setting. Ask for 8 conversions a second and each one is an average of about 31 000 internal samples. Ask for 860 and each one is an average of 290.

How many of the sixteen bits are real
128 SPS · ±2.048 V
Full-scale range
Data rate128 SPS
One conversion
7.8 ms
Noise, peak to peak
62.5 µV
Bits you can trust
16.00
At 128 SPS the last bit is real. The noise is 62.5 µV, which is exactly one count in this range — the converter is as quiet as its own resolution. One conversion takes 7.8 ms, and if you are reading a battery or a temperature that is nothing at all. Start here and only speed up when something forces you to.

The lower bars are every rate at the range you have selected, so the shape of the trade is visible at a glance: flat across the slow half, falling away at the top.

What TI actually measured

With the inputs shorted together — so the only thing being measured is the converter's own noise — the peak-to-peak figures come out like this at ±2.048 V:

RateNoise, peak to peakBits you can trust
8 SPS62.5 µV16
128 SPS62.5 µV16
250 SPS84 µV15.6
475 SPS79 µV15.7
860 SPS119 µV15.1

Up to 128 SPS the noise is exactly one count. The converter is as quiet as its own resolution, and the last bit means something. Above that it stops being true, and it gets worse in the narrow ranges: at ±0.256 V and 860 SPS, under fourteen bits are real.

You are still reading a sixteen-bit number. The last digits of it are just random.

Which to pick

Start at 8 or 16 SPS. A battery, a temperature, a water level, a load cell in a scale — none of these move fast enough to care, and the slowest settings are free precision. 8 SPS also lines up neatly with 50 and 60 Hz mains, which is part of why TI quotes its best rejection figures there.

Move up when something forces you to. Four channels read in turn at 8 SPS takes half a second per sweep. If that is too slow, 128 SPS costs nothing in bits and gets the sweep down to 31 ms.

Above 250 SPS, know what you are buying. It is the right corner for a current spike or a waveform, and the wrong one for anything you are going to put a unit on.

Single-shot, and the 150 µA

The chip powers up in single-shot mode, which is its default: it sits in a power-down state drawing about half a microamp, wakes in 25 µs when asked, converts once, and goes back to sleep. Continuous mode never sleeps and draws about 150 µA.

For a data logger taking a reading every minute, single-shot at 8 SPS is essentially free — 125 ms of work an hour. That is what the mode is for, and most libraries use it by default without mentioning it.

When it does not work

My readings wobble by several counts on a still voltage

Check the data rate before anything else. At 860 SPS in the narrowest range TI's own figures say the bottom two bits are noise, so several counts of movement is the chip working correctly. Drop to 8 or 16 SPS and the wobble usually disappears entirely.

I need it fast and quiet at the same time

Average in the sketch. Ten readings at 250 SPS take 40 ms and average down to roughly the noise of a much slower rate, at the cost of the I²C traffic. It is the same trick the chip does internally, moved into your code, and it is the standard answer when one rate will not do both jobs.

How long should I wait after asking for a reading?

One over the data rate, and no more — these conversions settle in a single cycle, so there is no extra time to allow. At 128 SPS that is 7.8 ms. Better than waiting at all is the ALERT pin in conversion-ready mode, which tells you the moment the number exists.

Is single-shot mode slower than continuous?

By about 25 microseconds, which is the time the chip takes to wake from its power-down state. In exchange it sits at half a microamp between readings instead of 150. On anything battery-powered that trade is always worth taking.

Where this goes next

The solder pad, and why TI's four addresses are two here.

Two addresses on this board

Edit this page — content/books/ads1115/data-rate-and-noise.mdx

Community

Questions about this product

See what other owners have asked, and read their solutions.

This page covers several products. Choose yours to see the right discussions.

Discuss this article

Ask about this page. The answer stays here, on the page it belongs to, for whoever hits the same wall next.

Browse Modules and blocks on the forum