Archive for category low noise design
Here’s the specs: Voltage noise 2.5nV/√Hz, Gain 1000, bandwidth 0.3Hz to 500kHz. I measure 600nVpp from DC to 1kHz at the output.
Here’s the story: Designing and debugging a high precision A/D stage, you will want to know how quiet the voltage reference really is. When I worked on a 20 bit A/D board, I found this amp a great way to prove that a simple voltage regulator was not good enough as a reference voltage (measured data trumps an any preconceived notions). It was a big help getting the shielding and grounding debugged too!
Since most ‘scopes have about 5mV/Div at their highest gain, this AC-coupled amp allows you to ‘see’ to 5uV/Division. For a 5V full scale A/D that’s 1ppm per division on your ‘scope. Trust me, you’ll see things there. (I read that Pasteur freaked out his dinner hosts, using his microscope to look for germs on his food. He was ‘debugging’ in his day too, with his new favorite toy).
Here’s a pdf of the schematic – it’s based upon the design in the book Low-Noise Electronic System Design, by Motchenbacher and Connelly (a highly recommended text). Since the text came without the amp, I had to make one myself.
I also have a SPICE model for it, using Linear Tech’s LT-SPICE and I’ll send it to you, if ask.
We’re working on a low-noise amplifier product, and to test it we need some attenuation, to generate clean low-level signals. The pdf shows a schematic and describes some test results. It’s a little more tricky than just a couple resistors because with every resistor you get (free!) parasitic capacitance that tends to distort your signals, unless you compensate the network.
Based upon Motchenbacher and Connelly’s excellent book ‘Low Noise Electronic System Design‘. Click on the link for the pdf or on the image for a JPEG version.
We like to measure things – and NIST is our source of the best measurement advice. We knew that time has been measured with astonishing accuracy, almost 1 part in 10^15.
Given how accurately time is measured, we wondered ‘how is the kilogram calibrated?’
We learned that, according to NIST,
“The magnitude of many of the units comprising the SI system of measurement, including most of those used in the measurement of electricity and light, are highly dependent upon the stability of a 131-year-old, golf ball-size cylinder of metal stored in a vault in France”.
And over the years the International Prototype Kilogram, the IPK, and its copies that are distributed around the globe, have ‘gained weight’. There’s more here at wiki: http://en.wikipedia.org/wiki/Kilogram
Here’s the graph to show it:
NIST further explains:
“Mass drift over time of national prototypes K21–K40, plus two of the IPK’s sister copies: K32 and K8(41).[Note 9] All mass changes are relative to the IPK. The initial 1889 starting-value offsets relative to the IPK have been nulled. The above are all relative measurements; no historical mass-measurement data is available to determine which of the prototypes has been most stable relative to an invariant of nature. There is the distinct possibility that all the prototypes gained mass over 100 years and that K21, K35, K40, and the IPK simply gained less than the others”.
Uncertainty, it seems, is here to stay.
However, there’s an improved kilogram standard being worked on, called the Watt Balance, which measures the electrical power used to null the weight of a one kilogram test specimen. To get accurate results, NIST must establish the gravitational force accurately. Here’s a link to a NIST article describing how they do it:
Which reminds me of a Jim Williams article from Linear Tech, where he describes a VERY accurate electronic weigh scale, capable of measuring your heartbeat – since all the blood pumping and flowing changes your weight, a little bit. He called it a ‘ballisto-cardiogram’. Here’s where you can read more, in AN43 that Jim wrote for Linear Tech, about bridge circuits – including how to make a scale that can resolve 0.01 pound at 300 pounds full scale, or about 33 parts per million. It uses a clever circuit to achieve balance quickly and accurately, check it out:
This presentation has a set of clear, well thought out images describing how chopper techniques can reduce 1/f noise, reduce drift, and even how to cancel the nasty charge injection of FET switches. It shows how modulation can reduce noise in a sensor amplifier system.
I first learned of Kofi Makinwa’s excellent work through the recent IEEE Solid State Circuits magazine, Winter 2010, Vol. 2, No. 1. He demonstrates a clever accelerometer that uses a small air volume as the ‘proof mass’. The Wheatstone bridge has been around a long time, but it’s clear it can be taught some new tricks. This is the first I’ve heard of a ‘nested chopper’ architecture. Great stuff. Check out Makinwa’s other publications at the IEEE.
I’ve spent some time trying to squeeze good data from MEMS sensors, and I know how difficult it can be. These articles show why adding some switches and circuit complexity can really pay off. And it’s only CMOS and FETs, so we get ‘em for free from Moore’s law, right?
We notice the assertion that A/D converter quantization noise is equal to ADU/SQRT(12), where ADU is the quantization unit or LSB. We saw this in Hobbs’ excellent book Building Electro-Optical Systems, Making It All Work.
So, we decided to derive this. Took us a while to get the ‘trick’, and to remember how to perform calculus, to get that pesky root-mean-squared function.
Think of the quatization error as a sawtooth function that repeats. Then work out the RMS noise of that sawtooth wave (it happens to be the same as a triangle wave). And, yes, it does work out to that value.
Now the next part is Hobbs’ assertion that this quantization noise is not a Gaussian distribution. Get to work.
Here’s a couple sources for precision resistors.
One source seems to provide a great price/performance trade, the other is really the best quality resistor you’ll be able to buy. BUT precision low drift resistor arrays require careful inspection of the data sheet. Here’s a couple examples.
Recently I noticed that Maxim IC, (famous makers of 5V powered RS232 interface chips), have begin selling a pair of precise resistors in a SOT23 package, at an attractive price.
This family of parts, the MAX5491, claims to have 2ppm/degC resistor to resistor drift of the pair in the same package. Take care – sometimes the large print giveth, and the small print taketh away – the absolute tempco is spec’d as 35ppm/degC, so if you compare two different individual SOT23 units with each other, they may drift more that that nice small 2ppm value at the top of the spec sheet!
For better matching, the VFCD1505 parts made by Vishay, also available at DigiKey, are spec’d for 0.2ppm/degC drift, or 10X better than the Maxim parts (and, well, about 5.7X the price, at $20/qnty 1, vs the $3.50 qnty 1 at DigiKey).
I like that Vishay also details subtle values such as ‘voltage coefficient’ and the current noise (or sometimes called ‘excess noise’, noise that is beyond the thermal noise of the resistor value).
The specifications of the Vishay parts are about where the state of the art is for actual resistors you can buy. Better resistors can be found only at NIST.
Careful consideration of all the elements of a system’s design can lead you to some very improved performance. Imagine improving a benchtop NMR system by making it 60 times lighter (120kg to 2kg), 40 times smaller, and yet 60 times more sensitive!
This article, from the IEEE Journal of Solid State Circuits (Vol. 44, No. 5, May 2009), shows an excellent example of how this occurs.
link to IEEE abstract of ‘CMOS RF Biosensor Utilizing Nuclear Magnetic Resonance’ by Sun, Liu, Lee, Weissleder, and Ham
I recommend reading the article – it’s very well written, it describes how NMR works, and it details their systems approach to their improved design. Much can be learned here. The use of a resonant circuit for gain (they call it ‘passive amplification’) is detailed in Figure 8 of the article. (It reminded me of the old ‘regenerative’ type radio receivers, back when a vacuum tube had a power gain of about 12).
Put another way, this article shows that the ‘building block’ approach, when off-the-shelf 50 Ohm compatible RF modules are used, makes it easy to build a system that works – but that it leaves out some great performance improvements that are only possible when you analyze the basic system operation and theory. The design improves when you ask questions like ‘why 50 Ohms’ or ‘where does that noise originate and how can I maximize the signal’ and ‘how can I make this work with a much smaller and lighter magnet’? The article also answers ‘now that I can use a small magnet, can I make a custom CMOS IC that performs the RF detection, and seriously reduce system cost and size’?
Buying as much stuff off the shelf is not bad – it’s a great way to get a proof of principle working FAST, and it demonstrates that an idea or technique can work. Nothing says ‘success’ like working hardware – it allows the investors, managers and engineers to breathe easier.
But that extra performance gain from really digging into the details of how things work can pay off – in this case, it changes a benchtop lab instrument into a battery operated portable clinical test platform – this opens new opportunities and situations where this NMR system can be utilized.