# Re: Minimal Detectable Signal

1. Jun 8, 2007

### Michael Lin

Hi all,

I want to detect a 50 uV (microvolts) signal from a sensor. I was wondering what's the minimal detectable signal from the circuitry on the market/in the industry now? I know that if I have a humongous setup for noise supression, I can detect 1 uV even, but I do have a size limitation on the amplifying and conditioning circuitry. In short, it has to be small (less than 1cm by 1cm). Does it sound impossilbe to detect the 50 uV signal with the space limitation?

Thanks,
Michael

2. Jun 8, 2007

### waht

What exactly are you trying to detect? A DC voltage or an RF signal?

In either case it's more than feasible to detect 50 uV.

3. Jun 11, 2007

### xez

50uV is generally a trivially easily detectable signal level
with respect to its magnitude.
The complications will only be relative to :

1) the fidelity and precision at which you need to know the
sensor's signal. e.g say 50uV is the average amplitude, it's
possible that you'll want much better resolution so you
can know not only whether 50uV is "present or absent",
but to have a very well resolved waveform of the 50uV
magnitude signal's waveform to see the fine details of that
waveform. So therefore you'd need perhaps much better
than 50uV resolution to have a quality picture of the
DETAILS of the 50uV amplitude signal waveform.

2) The quantity and type of noise present that is in
the same frequency range as the signal to be measured.
If there's a lot of noise intrinsic in the sensor, in the
environment, in the measurement circuitry, then your
measurements will have some uncertainty and corruption
due to the noise. The figure of merit is called the
"signal to noise" ratio and it determines the achievable
accuracy and confidence level of any particular
measurement. Using averaging on multiple measurements
can help reduce the uncertainty due to some kinds of
random noise if the signal being measured has some
correlation / statistical properties that is different than
the perhaps more 'random' (or at least distinct) types of
noise.

3) The frequency bandwidth of the signal of interest.
It's more challenging to measure extremely wide
(or extremely narrow) bands of signal frequencies in the
presence of noise and relative to the constraints of your
measurement circuitry than it is to measure moderately
narrowband and low frequency types of signals.

For example, a very common and cheap
16 bit analog to digital converter (which is basically
common in even $25 computer "sound cards") will measure the minimum signal of 1/65536th of its maximum allowable signal level. If the maximum allowed signal level was 2.5 Volts (a fairly common value) can resolve a minimum of a 2.5 Volts / 65536 = 38 microvolt signal at its least significant bit, so it'd tell the presence or absence of a 50uV signal but you'd have no real resolution of the details (waveform) of the signal. Fortunately, it's also very cheap and trivial these days to get even 20 or 24 bit ADCs (analog to digital) converters which can measure at resolutions of 1/2^NBITS of the applied maximum voltage reference values, so... 20 bits ADC: 1/2^20 = 1/1048576 = 0.95 microvolts minimum resolution per volt of the full scale reference voltage. So with Vref = 2.5V that's 2.5 * 0.95micro = 2.375 microvolts being the minimally resolved signal. Relative to a 50uV PEAK sensor signal, a resolution of 2.375uV would give 50uV/2.375uV a resolution of around 21 'steps' of sensor signal waveform shape information with very common and inexpensive measurement circuitry. Typically 20 bit ADCs would be available to run up to a sampling rate of at least 96 kilohertz, so in measuring the sensor value at 96kHz rate you could discern sensor fluctuations in a bandwidth of well less than 48kHz or so. It's also very common to use integrated circuit amplifiers to preamplify the input sensor signal before that signal is fed to an analog to digital converter to be measured. I've recently done some consulting on the design of a project that used a common$3 dual op-amp with a gain
of 40,000 to amplify a 1 microvolt level antenna sensor
signal up to the level of 40 millivolts (x40000) prior
to that greatly amplified signal being fed to a
10 bit Analog to digital converter chip to be measured.

level signals by factors of something like ten million
before they're powerful enough to be decoded and used to
power speakers etc.

As to what can be achieved in 1cm^2 of PCB real estate?
Well you can certainly fit a low noise op-amp to amplify
the signal and something like an SPI 12bit ADC and a
small microprocessor (PIC or AVR type) in such a space.
If your microprocessor was elsewhere and you only needed
to digitize the signal with an op-amp and ADC and feed
it to a connector then that would be even easier and
could be done with one or two common cheap surface
mount ICs.

Thermocouples generate only a few of microvolts of
signal per degree centigrade and single chip thermocouple
digitizer chips just a few square millimeters commonly
measure those signals with a precision of a millivolt or two
for around $2/chip. Similar chips are used to monitor the charge and discharge functions of lithium ion batteries and they deal with measuring comparably small voltages. http://www.linear.com/pc/productDetail.jsp?navId=H0,C1,C1155,C1001,C1152,P37907 LTC2450 - Easy-to-Use, Ultra-Tiny 16-Bit Delta Sigma ADC * GND to VCC Single-Ended Input Range * 0.02LSB RMS Noise * 2LSB INL, No Missing Codes * 2LSB Offset Error * 4LSB Full-Scale Error * Single Conversion Settling Time for Multiplexed Applications * Single Cycle Operation with Auto Shutdown * 350µA Supply Current * 50nA Sleep Current * 30 Conversions Per Second * Internal Oscillator—No External Components Required * Single Supply, 2.7V to 5.5V Operation * SPI Interface * Ultra-Tiny 2mm x 2mm DFN Package That'll measure around 1/2^14th of a 3V signal, so around 183uV minimum detected signal in a tiny space, so if you preceded it by an appropriate low noise op-amp with a voltage gain of 500 that'd turn your 50uV signal into a 25mV signal and then that'd be digitized to a resolution of 25mV/183uV = 136 steps of resolution, all within your space requirements for well under$20 worth of parts and printed circuit board.

As the other poster mentioned, Maxim has several
16 and 18 bit ADC chips in the required size, sensitivity,
range. e.g.
http://para.maxim-ic.com/cache/en/results/4942.html

www.ti.com has several low noise wide bandwidth
good linearity op-amps in small packages, some
capable of single 3.3V supply operation, for well under
\$5 single quantity; digikey.com or mouser.com or other
such electronic parts distributors will probably sell all
the parts you need for a prototype in quantity one or more.
Maxim, TI, et. al. typically also sell "evaluation boards"
for their ADC and OP-AMP et. al. parts, they'll be physically
larger than you need, but may be a convenience for
quick "lab bench" experimentation before you build
custom production PCBs that are of small size.

4. Jun 12, 2007

### edmondng

look up on 'Lock-In Amplifier' or PSD. I myself am trying to measure really small signal, just integrate it over time

5. Jun 12, 2007

### Staff: Mentor

I think these are the key points. To get a good SNR measurement, you need to limit the bandwidth of the measurement as much as possible, to eliminate noise that is not in the same frequency range as your signal. That may mean using a sharp BPF and lock-in amplifier for a narrow-band AC signal, or maybe a steep LPF for a DC signal, etc.