Let's made a PC-base metal detector with usb interface !!!

[Aziz]

Hi all,

I have started again to work on transistor matching. My DMM is not accurate enough (only 4000 counts). But hell, why not using the sound card for much more dynamic range (six decades!)?

I will work on a test board, which will make it elegantly. And this on a test board destined for differential pair with the intended operating collector currents. It can be used to match resistor pairs too. Of course, noise behaviour can also be measured. So I can try different transistors for low noise operation.

The tricky solution: CMRR analysis (with shorted differential inputs to ground). More on later.

The thermal coupling of the transistors must be done through the leads of the transistors. The plastic case is a bad heat conductor. But the collector lead is going directly to the chip die. The base and emitter leads are not good enough (wired with a very thin gold wire to the chip die). Anyway, they conduct the heat to the plastic case good.


Aziz

[Aziz]

Hi all,

I have built the mentioned test board (differential pair). It seems to be a very fine transistor matching system (hFE and Ube) for NPN and PNP transistors (just exchanging the power supply polarity). I can get 4-5 digits matching accuracy with the sound card measurement using the 24-bit resolution. More on later.

But I have found, that the BC327-40 (PNP)/BC337-40 (NPN) transistors work more quiet than the BC550C (NPN)/BC560C (PNP) variants. 3 dB lower noise figure (BC327-40 tested)! This is quite surprising, because there are no noise figure (NF) specifications for them yet.

My measurements confirm the benefit of BC327-40/BC337-40 mentioned in the following paper:
http://www.janascard.cz/PDF/Ultra%20...amplifiers.pdf

Really funny thing. I will buy some super-matched transistor pairs next month to compare them to the hand-matched types.

Aziz

[WM6]

Originally Posted by Aziz

But I have found, that the BC327-40 (PNP)/BC337-40 (NPN) transistors work more quiet than the BC550C (NPN)/BC560C (PNP) variants. 3 dB lower noise figure (BC327-40 tested)! This is quite surprising, because there are no noise figure (NF) specifications for them yet.

Thanks for interesting info Aziz. Enjoy your work.

[Aziz]

Hi all,

the BC337-40 (NPN) is also tested now. It achieves very good noise performance as well (at collector current of 1.2 mA).

Both outputs of the differential pair stage were fed into the sound card input line (left and right channel). The difference is made digitally of course. My noise floor level is at -114 dB (no input plug connected to the sound card). When I plug the test board, the noise floor increases only by 1 dB into -113 dB.
There is a signal level distance of 4-5 dB to the BC550C pair (noise floor at -108 dB).

So the BC337-40 and BC327-40 are working outstanding. I will order some more samples of these transistors soon (I had approx. 15 pieces of each already and found a good match in the batch).

Aziz

[Aziz]

Hi all,

here is the differential-pair transistor matcher. I have found this proposal somewhere in the internet forums.

First try to find two same 2.2k collector resistors (measure this with your DMM). Their matching accuracy should be better than 1%. Otherwise the offset voltage might be too much and the operating point moves into the saturation region of the transistors. If you have very accurate collector resistors, R9,R10 and U1 can be dropped.

Select the collector current (no signal fed into the inp). The pot U2 is for adjusting this. You can measure the collector current if you measure the voltage drop across R1 or R2.
Collector current = UdropR1/R1 = UdropR1/2200
For most applications, a collector current in the range of 1 mA is ok. So a voltage of 2.2V across the resistor R1 should be there.

Feed an AC signal into the inp input (1 kHz sine wave, Sound card/function generator). If both transistors Q1 and Q2 are well matched, the AC and DC voltage across the transistor collector will be low. The lower, the better matched. The DC voltage is the amplified offset voltage mismatch (Vbe mismatch). The offset voltage gain is around x80 I think (spice simulation).

Matching strategies:
1. You can compare REF against DUT (minimising the measured DC/AC voltages).

2. Or you can change the DUT's and note the measured DC/AC voltages. The two almost identical measured DUT's do match together.

Both transistors need a thermo coupling. If the temperature is differing, DC and AC voltages will heavily change. I recommend to place the whole circuit in a box during measurement. After both transistors get same temperature, the measurement can be done.

You can use any DMM for DC/AC voltage measurements. Or you can use the sound card.

Aziz

[Aziz]

Hi all,

I have just ordered another bunch of interesting parts. This time two super-matched transistor pairs, which will (hopefully) be available:
LM394 (NPN)
MAT03 (PNP)
(no MAT02's there... )


I don't know, whether the hand-matched transistor pairs will be better (quiet). So I have ordered a batch of:
100x BC337-40
100x BC327-40

100x 2SA1015
100x 2SC1815
Let's look, how the Japanese transistors work. They claim to have a typical noise figure of 1 dB. Can they outperform the BC337/BC327? And how the super-matched will work? Well, I will see this soon.

At least, I will have a life-time stock of these transistors above.

Aziz

[WM6]

Originally Posted by Aziz

At least, I will have a life-time stock of these transistors above.

Aziz

Hi Aziz,
where you decided to order?

[Aziz]

Originally Posted by WM6

Hi Aziz,
where you decided to order?

here:

http://www.reichelt.de

Aziz

[Aziz]

Hi all,

look at this obsolete part: µA726
(see below data sheet)

I even have been thinking of doing almost the same. But in a different and state of the art variant.

Aziz

[Aziz]

Hi all,

the new parts just arrived and I am very happy to have them.
Now the very interesting questions are awaiting to be answered.

It's really very difficult to match the transistors due to ambient temperature changes. I probably will build a temperature controller to keep them at specified temperature. I also have ordered some temperature sensors (LM335) for this purpose and for the PI Controller.

A lot of important parts in the PI Controller are temperature sensitive. I will track them and process the tempco in the software.

Let's go..

Aziz

[Aziz]

Hi all,

I have a first result of the noise performance of the transistor-pairs now. For this measurement, I have modified the test circuit into a high gain single-ended DC amplifier. The gain is set to 60 dB (x1000). This makes noise comparisons of the transistors more accurate. The input is grounded, the output is observed in the FFT spectrum (up to 45 kHz) and the noise floor level is measured around 10-20 kHz.

Sound card noise floor level (signel ended): -116 dB (left channel, input plug not connected)

Transistor-pair, Noise floor level
========================
BC550C, -85.5 dB
2SC1815-Y, -89.5 dB
BC337-40 (from new batch), -94.5 dB
BC337-40 (from old batch), -97 dB
LM394-CH, -97 dB

The highest noise has BC550C pair, the lowest noise the LM394. Interestingly, the BC337 worked as well. The new batch has slightly higher noise. It probably depends on the manufacturer.
But the LM394 worked outstanding stable. The hand matched pairs had offset voltage drifts due to thermal grandients. For an AC amplifier, this should not matter much. But for a high gain DC amplifier, it is a K.O. criteria.

I was dissappointed from the 2SC1815. The hFE classification Y (yellow) has the lowest hFE (around 200) and this was not specified from the supplier. To have a high gain transistor is necessary to lower the bias current. My LM394 has hFE well above 1000 (around 1200). The LM394 is extremely temperature stable.

What's the next? I will buy more super-matched transistor pairs in the coming month.

The voltage controlled amplifier (VCA) for the PI Controller project is getting more realizeable now.

Aziz

[Aziz]

Hi all,

I am really not very amused about the VCA performance. Spice simulations predict high noise contribution to the signal chain at low gains (x1 .. x20). If I compare it to the common products (VCA610, VCA810, ..), they also behave similar. No a real benefit. I even have not built it yet.
Ok, it's time to kill the VCA part.

PGA? Don't know yet. Very likely trashing the idea too.
I have a much better solution with the lowest noise performance.

I am looking into some MUX/analog switch chips now. I don't like their huge non-linear behaviour at +/-5V. I have to rise the supply voltage. Fortunately, they need not much power. Yet another charge pump is beeing designed. This time much better, symmetric output, synchronized, low noise, different voltage levels.
The CD4051..4053 can be powered from +7.5/-7.5V
The MAX312..MAX314 can be powered from +15/-15V.
Now it was a good idea to have a 8.5V system voltage. All relevant supply voltage levels can be generated from the 8.5V system voltage with less circuit overhead.

Aziz

[Aziz]

Fnckn!

Sorry, but LTSpice had bad noise calculations on their own op-amp model (LT1007). Something got really wrong. I have recognized this with a simple closed-loop differential-pair amplifier simulation.

I know, do not use their mosfet models.
Do not use their diode models.
Now, do not use their op-amp models too.

Now, this seems to be very interesting. Noise contribution has been dramatically reduced... Something, I can start optimizing the VCA.

VCA is not died yet.

Aziz

[Aziz]

Hi all,

now the VCA is getting very very attractive. Two MAT02 or LM394 are forming a good simple VCA with 1nV/sqrt(Hz) voltage noise density at maximum gain. And 3nV/sqrt(Hz) at minimum gain, in which the output stage is more dominating here. The results were obtained making a spice circuit simulation.

To get the low noise operation, I had to remove all base resistances (input voltage divider). This reduces the input voltage range of the VCA (max 50 mV) for linear operation and fortunately increases the input impedance. Anyway, we do have very low signals and we need not to attenuate the input signal. For very early signal sampling, the pre-amp output should be used. The VCA is versatile configurable for variable gain and linear operation region.

And I was coming near to trash the VCA. But the circuit needs to be optimized further.

Aziz

[simonbaker]

We appreciate your hard work Aziz -- it may come in handy in many different ways.

-SB

[WM6]

Originally Posted by Aziz

The software part will do all the things required for the operation:
- pulse timing, pulse width, synchronisation, pulse operation mode (all with stereo line output: left/right output channel)
- frequency domain analysis (FFT/Lock-in amplifier/Goertzel algorithm)
- signal analysis and feature extraction (discrimination, filtering, ground balancing, ..)
- detection signalling (sound output in the internal sound card of the netbook)



Aziz

Hi Aziz, what do you mean about use Anisotropic filtering in deeep signal extraction?

http://en.wikipedia.org/wiki/Anisotropic_filtering

[Aziz]

Hi WM6,

Originally Posted by WM6

Hi Aziz, what do you mean about use Anisotropic filtering in deeep signal extraction?

http://en.wikipedia.org/wiki/Anisotropic_filtering

this technique is totally different from our application. We don't really need it.

There are lots of other digital signal processing algorithms to meet our requirements. For deep signal extraction, digital lock-in amp or Goertzel algorithm is the best. Particularly for decoding our signal channels coming from the sound-card line.

They have excellent small bandwidth (adjustable) in which the noise is reduced almost into its noise spectrum density. Due to this fact, I can decode several signal channels on a single signal line. The channel seperation is really excellent and there is (almost) no crosstalk between them.
The two above mentioned signal decoding algorithms also have a built-in low pass filtering and box-car integration technique given by their time constant parameter.

Just an example:
If I have a 1 V rms dynamic range on my 24 bit sound-card input, I am able to decode 1 µV rms signal levels (even lower with higher time constants). This results in a 20*log10(1V/1µV) = 120 dB dynamic range or ld(1V/1µV) = 20 bit digital resolution.

If my sound-card has a SNR of let's say 110 dB (-110 dB noise floor), the noise level reaches 10^(-110/20) = 3.2 µV rms. So I have to go well beyond -120 dB to detect the 1µV signal savely. Without any signal processing algorithms, I can't do it as the sound-card wouldn't allow it (-110 dB noise floor).

Aziz

[Aziz]

Hi all,

it is a pure waste of money to put an expensive super-matched transistor pair for a current mirror used in the VCA. There are much cheaper matched transistors available just for a current mirror application:
BCV 61 (NPN) and BCV 62 (PNP). I have already bought several of them in my last order. Well, it's a SMD part but can be soldered in a precision 8-pin DIP IC socket easily.

So I can start testing the VCA soon. VCA application is very critical as the controlling voltage must be very clean and stable. Otherwise, it would modulate the amplified signal and would result in producing a "modulated noise".

Aziz

[Aziz]

Hi all,

I have improved my VCA a bit: Now
- fully differential input
- fully balanced output (symmetric)
- automatic offset voltage compensation (no adjustments necessary)
- lower total noise

Well, I have used total:
- 3 current mirrors (2x PNP type, 1x NPN type)
- 1 differential-pair transistor (MAT02/LM394)
- 3 op-amps (2x transimpedance, 1x servo control)

At the end, the number of op-amps remains same. I would need anyway a buffer and an inverter to make a balanced output signal for further processing. And one for the output stage of the VCA of course.

But this version has lower total noise as these op-amps are used at the output-stage of the VCA. And offset voltage adjustment: gone!

The temperature dependent gain (gm = Ic/VT, gm=Transconductance, Ic=Collector Voltage, VT=Thermal Voltage) will be compensated via the software (temperature sensor).

Aziz

[Aziz]

Hi all,

this is the frequency response of the amplifier (below). I even had to compensate it to reduce the bandwidth. Collector current of differential-pair transistor is max. 2 mA (high gain). The input voltage noise spectrum density even gets better. Not bad yet.

Aziz

[Aziz]

Hi all,

the VCA concept is not convincing me as there is a much simpler, cheaper and better solution. No, it's not a PGA. The PGA-switches in the feedback loop path would contribute much noise to the amplifier. Not good. PGA-idea is dead.

But the concept of having several amplifiers is much better regard to stability, noise and temperature dependency.

The signal path would look like this:

Pre-amp -> Amp1 -> MUX
Pre-amp -> Amp2 -> MUX
Pre-amp -> Amp3 -> MUX
Pre-amp -> Amp4 -> MUX
.. more amplifiers, if needed ..

The amplifier stages should have a fixed gain factor (Gain1, .., Gain4). And log(Gain1/Gain2) ~= log(Gain2/Gain3) ~= log(Gain3/Gain4) behaviour (exponential gain factors). Four amplifier stages should be really enough to have a flexible headroom in the time domain range (early and late sampling timings). No doubt, the VCA would give a finer dynamic range possibility, but it's not really necessary.

After the multiplexer, the signal would be buffered and inverted:
MUX -> buffer (signal+)
buffer -> inverter (signal-)

It is very important, that the input impedance of the buffer should be not low. Otherwise, the MUX itself would give more temperature dependency to the system due to input bias currents. Therefore, the low input impedance inverter is following the buffer.

This concept is making the DAC gain control voltage for the VCA also obsolete. Only 2/3 bit control lines are necessary to switch the MUX. Another plus: very fast switching between different gains possible. A VCA would introduce a settling time between gain changes.

It's just simple and stupid. But it would have the best performance.
KISS principle.

Aziz

[WM6]

Originally Posted by Aziz

Working on relative changes instead of absolute phase angles, gives a better and easier implementation.

Thank you Aziz, but can be those usable, mean relialble enough in repeatability to give us correct info?

[Aziz]

Originally Posted by WM6

Thank you Aziz, but can be those usable, mean relialble enough in repeatability to give us correct info?

I don't know, which posting number you are referring to.
Can you give me the context please?
Aziz

[WM6]

Originally Posted by Aziz

I don't know, which posting number you are referring to.
Can you give me the context please?
Aziz

Sorry, by mistake from the end of first page, post #35. So crauts live everywhere. But are you sure about FPGA? You ca set multiple amplifiers input inside PGA too?

[Aziz]

Originally Posted by WM6

Sorry, by mistake from the end of first page, post #35. So crauts live everywhere. But are you sure about FPGA? You ca set multiple amplifiers input inside PGA too?

Sorry, can't find the context. Can you give the referring absolute post number? Then your questions.
Aziz