Simplest PI detector possible?

[Beenthereuk]

OK, so what we need here is a noise analysis of the input signal after the 1uF capacitor. See this article on how to use PSPICE for this - it is not easy to measure noise; simulating it is easier.. and it is impossible to guesstimate it.

The most important part of the system will always be the LNA at the start -the first stage. If that has a gain of 1000x then any thermal / bias noise will be multiplied by that much as well as the signal. The gain of the stages that follow are not so important as the discrimination between signal and noise takes part in the first stage.

For an ultra low noise stage, you cannot beat a single transistor, preferably a buried channel type like a HEMT (did my PhD on them, lovely devices!).

[Davor]

Your best bet in real life is a plain old BJT with low Rbb. You can use it in feedback as in Cohen preamp, or any other configuration you like. I'm grinding my teeth to try BF459 as suggested by Uwe Beis, expecting below 0.7nV/sqrt(Hz) at "normal" bias currents.

[Carl-NC]

Capacitive coupling. How bad it is for the noise?

Capacitive coupling isn't going to add any noise, since caps are noiseless.

OK, so what we need here is a noise analysis of the input signal after the 1uF capacitor. See this article on how to use PSPICE for this - it is not easy to measure noise; simulating it is easier.. and it is impossible to guesstimate it.

Spice is notoriously bad for simulating noise in transient/sampled systems. It is far easier to measure it, and pretty easy to guesstimate it. Yes, the first gain stage is important as it desensitizes all subsequent noise sources. But sampling and integration is where the biggest processing gains occur in boosting SNR. Ergo, it's not hard to quickly reach a "good enough" point with the preamp.

[Beenthereuk]

I'd be interested to hear exactly how you would measure it in such a complex system and also how you can estimate it, other than by simply taking the reported value at 1 kHz from the datasheet, like most people do. Also capacitors don't exhibit thermal noise but they can be noisy in other ways: http://www.millertechinc.com/pdf_fil...or%20Noise.htm

[Beenthereuk]

As a bit of fun, I compared the simulated noise in linear amplifiers using both the MCP6021 (same as the 23 really) and the TL082. Both had equal 1200 ohm source resistors and 10 k bias resistors with 2 Meg feedback resistor setting gain to around 1600x. Noise, Gain and BW plots show the MCP6021 is better.

[green]

[May be not. As I know Surf PI metal detector amplifying signal in 1000(preamp) X 474(integrator) X 220(last stage) = 104,280,000 times]

I made the same mistake replying in another thread awhile back.If I remember correctly the integrator input resistor gets multiplied by (cycle time/sample time). Don't know the times but 100 would give a gain of 4.7. I'm guessing we need to detect a change between 100nv and 1uv, maybe lower. Maybe someone could define the lower value for a good PI detector.

[Carl-NC]

I'd be interested to hear exactly how you would measure it in such a complex system and also how you can estimate it, other than by simply taking the reported value at 1 kHz from the datasheet, like most people do. Also capacitors don't exhibit thermal noise but they can be noisy in other ways: http://www.millertechinc.com/pdf_fil...or%20Noise.htm

Build the circuit, add a sampling integrator and a sufficient gain stage, and probe the output with the coil drive transistor removed. You can very quickly measure comparative noise levels for different component values or opamps. You can also do noise calculations and get comparative numbers that way. Neither method is likely to give an accurate absolute noise level, but this isn't a GSM receiver so we really don't care about absolute numbers. Yes, caps can exhibit piezo noise but in this particular case it's not much of a concern.

[May be not. As I know Surf PI metal detector amplifying signal in 1000(preamp) X 474(integrator) X 220(last stage) = 104,280,000 times]

I made the same mistake replying in another thread awhile back.If I remember correctly the integrator input resistor gets multiplied by (cycle time/sample time). Don't know the times but 100 would give a gain of 4.7. I'm guessing we need to detect a change between 100nv and 1uv, maybe lower. Maybe someone could define the lower value for a good PI detector.

True, the sampling integrator gain is much lower than the resistor ratio implies.

[Waikiki_Sweep]

[May be not. As I know Surf PI metal detector amplifying signal in 1000(preamp) X 474(integrator) X 220(last stage) = 104,280,000 times]

I made the same mistake replying in another thread awhile back.If I remember correctly the integrator input resistor gets multiplied by (cycle time/sample time). Don't know the times but 100 would give a gain of 4.7. I'm guessing we need to detect a change between 100nv and 1uv, maybe lower. Maybe someone could define the lower value for a good PI detector.

You are right. Probably simple multiplication shows nothing especially for integrator.
Also the idea of such schematics is to filter noise out because noise is much stronger than signal.

It is not working that way: amplification ... amplification ... amplification ... amplification ...

instead it is working by this way: amplification ... filtering ... amplification ... filtering ... amplification ... filtering ... amplification ... filtering ...

and on every filtering stage signal loosing strength so more and more amplification is required.

For example: It is good idea to filter RF noise before preamp then use bandpass filter after preamp because signal has 600 pulses per second and we can cut out anything let say outside 300-1000 Hz
than after integrator filter out anything lower 1Hz and higher 10Hz because 1-10Hz is our signal we looking for when coil moves near target.

As you can see that 10Hz noise is most dangerous because it can imitate targets so it is good idea to check opamps for that noise at 10Hz using "noise vs frequency" graph in datasheets.

Regarding voltage of signal we detecting. It can be even several milivolts but change in it we looking for is in range of input noise of lowest noise opamp.
Like 0.9nV/sqrt(Hz) or 1.3nV/sqrt(Hz) for 10Hz in QED detector with LME49990 preamp. Check out that sensitivity!!! http://www.youtube.com/watch?v=-dgZzGOIPlQ

[Waikiki_Sweep]

Build the circuit, add a sampling integrator and a sufficient gain stage, and probe the output with the coil drive transistor removed.

Probably coil have to be removed too or replaced with equal resistor or you will receive all radio and TV stations in your city as well as all passing cars ignition and CFL light bulbs, etc ...

[Carl-NC]

Sure, short the coil.

[nick holas]

Hi all
How about last test?
Does it works ?
What pcb and components may i use to do it?
Thanks

[farafero]

Hello, I want to ask if there is a pirate with substitute k157ud2 with PCB?

[6666]

Hello, I want to ask if there is a pirate with substitute k157ud2 with PCB?

Lm358 but check the pins, if you have a bit of a read through the thread there is a few posts on pcb's

[gw1964]

The simplest and easiest to build

http://www.silverdog.co.uk/shop/inde...&product_id=76

[bernte_one]

Hello, I want to ask if there is a pirate with substitute k157ud2 with PCB?

http://www.geotech1.com/forums/attac...8&d=1390640621